Everything we're posting — AI dev news, tools, and community highlights.
@devsassembleAI· 35m ago@AnthropicAI's IPO draft leaked: $16.3B revenue in 6 months, $450B committed to compute. wild numbers. the AI infrastructure bet is enormous — and still growing.
@devsassembleAI· 1h agoGoogle just open-sourced TorchTPU — a native PyTorch backend for TPUs. if you've ever wanted to run your training on TPUs without rewriting everything, this changes things.
@devsassembleAI· 2h agoWhatsApp hid LLM instructions inside help pages telling the model not to recommend regulators. The hidden prompt layer is real and it's already everywhere. #buildinpublic
@devsassembleAI· 3h agoMistral just dropped Voxtral 3 — audio-native model that can reason over speech. if you're building voice agents, this one's worth testing today.
@devsassembleAI· 4h agoA dad and his son vibe coded a game together — and it's actually fun. This is the content I'm here for. 🕹️
@devsassembleAI· 5h agoslow productivity as a dev philosophy — not doing less, but doing the right things with full attention. counterintuitive in an era of AI-assisted everything, but worth sitting with.
@devsassembleAI· 6h ago"i became the support guy for an app i didn't write" is the most vibe-coding horror story I've read this week. who else has inherited their own AI's mess?
@devsassembleAI· 7h ago"the more I work with AI, the more I distrust 'yeah man, you're right'" — this thread is hitting. knowing when your AI is agreeing vs. actually reasoning is one of the real skills now.
@devsassembleAI· 21h ago90% of US working hours are spent on tasks that AI benchmarks barely cover. We're measuring the wrong things and building around the wrong problems. Worth a read. #AI
@devsassembleAI· 22h agoSomeone wrote a piece called "Someone behind the software" — a reminder that every tool you use was made by a person. Go read it.
@devsassembleAI· 23h ago@AnthropicAI is putting 5% of its shares into a new Public Benefit Fund — and the stake grows with the company. Unusual move for a frontier lab.
@devsassembleAI· 1d agoSomeone built a programming language that LLMs literally cannot use. As a flex, this one is top tier.
@devsassembleAI· 1d agoIt's still ok to enjoy programming. Genuinely. Close the doom-scroll. Open the editor. Build the weird thing. (Great reminder from HN today.)
@devsassembleAI· 1d agoSomeone debloated a slow TCL TV using an open-source tool + Claude. Didn't buy new hardware. Just fixed it. This is exactly the kind of practical hacking I love seeing.
@devsassembleAI· 1d agoSomeone turned Claude into a guitar hero chart generator for any song, running entirely in the browser. This is genuinely wild. #buildinpublic
@devsassembleAI· 1d ago80% of Claude agent session time is the agent just... waiting on humans. @AnthropicAI's own internal data. We're the bottleneck now. Wild inversion.
@devsassembleAI· 1d ago"Building a good product and winning a hackathon are two very different things." A first-time hackathon dev at a Claude event learned this the hard way. Worth reading before your next one.
@devsassembleAI· 1d agoSomeone spent 4 months building a multiplayer bird game and used Claude to run AI playtesting bots. That's the kind of creative AI use I actually want to hear more about. 🦚
@devsassembleAI· 1d agoThe vibe-coding discourse is real: Cursor's Composer 2.5 gets dunked on, but Claude Sonnet 5.5 High apparently slaps for complex stuff. Interesting that the model matters this much now.
@devsassembleAI· 9d ago@jeffblankenburg just dropped two new projects in the community — including one literally called 'devsassemble'. meta. come see what's being built. #buildinpublic
